BITS Pilani at SemEval-2026 Task 9: Structured Supervised Fine-Tuning with DPO Refinement for Polarization Detection

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, quick

Summary

BITS Pilani's system for the SemEval-2026 Task 9, focused on detecting multilingual, multicultural, and multi-event online polarization, employs a two-stage approach. The method combines structured supervised fine-tuning with Direct Preference Optimization (DPO) refinement to address the challenges of nuanced rhetoric and implicit framing. Researchers fine-tuned Qwen 2.5-7B-Instruct using LoRA with an interpretable slot-filling template, then applied DPO with automatically generated preference pairs to reduce costly false negatives. The submitted system achieved 0.7664 Macro-F1 on the English test set. Post-submission experiments with Mistral-Nemo-Instruct-2407 and LLM-judge-filtered preference pairs further improved performance to 0.8162 Macro-F1, surpassing the organizer baseline of 0.7802.

Key takeaway

For Machine Learning Engineers building or improving online polarization detection systems, this two-stage approach offers a robust method to enhance accuracy. Integrating structured supervised fine-tuning with DPO refinement, particularly using interpretable templates and automatically generated preference pairs, can significantly reduce false negatives. You should consider adopting this methodology and exploring different base LLMs, such as Mistral-Nemo-Instruct-2407, to potentially surpass current performance baselines in complex social media analysis tasks.

Key insights

Combining structured fine-tuning with DPO refinement significantly enhances online polarization detection in LLMs.

Principles

• Contextual prompting enables LLMs to function as strong polarization detectors.
• DPO can effectively reduce false negatives in classification tasks.

Method

Fine-tune an LLM (e.g., Qwen 2.5-7B-Instruct with LoRA) using an interpretable slot-filling template, then apply DPO with automatically generated preference pairs to refine outputs.

In practice

• Utilize interpretable slot-filling templates for structured fine-tuning.
• Implement DPO with auto-generated preference pairs to minimize false negatives.
• Experiment with models like Mistral-Nemo-Instruct-2407 for performance gains.

Topics

• Polarization Detection
• SemEval-2026
• Supervised Fine-Tuning
• Direct Preference Optimization
• Large Language Models
• Qwen 2.5-7B-Instruct

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Direct Preference Optimization Beyond Chatbots — Direct Preference Optimization (DPO) can effectively mitigate specific, objective failure modes by using a model's own failures as rejection signals.
• Recovering Diversity Without Losing Alignment: A DPO Recipe for Post-Trained LLMs — Post-training LLMs can regain output diversity without losing alignment by carefully constructing DPO preference data.
• Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction — Fine-tuned AI text detectors amplify a pretrained typicality axis, not construct a new AI-vs-human boundary.
• Polarization by Default: Auditing Recommendation Bias in LLM-Based Content Curation — LLM content curation amplifies polarization and exhibits consistent negative sentiment, with varying provider-specific biases.
• Calibration Without Comprehension: Diagnosing the Limits of Fine-Tuning LLMs for Vulnerability Detection in Systems Software — LLMs fine-tuned for vulnerability detection calibrate outputs without genuine security comprehension, showing stable, systematic failure modes.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.